WGCNA Workflow from GEO FASTQ Files – Expression Type, Sample Inclusion, and Preprocessing

[tutorwhizzz]

Hello, I’m new to bioinformatics and would appreciate some guidance on the general workflow for WGCNA analysis in disease studies. If there are any tutorials or resources you can point me to, that would be appreciated! I watched the tutorial from Bioinformagician, but she only performs WGCNA using count data.

Expression data: What type of expression data is best for WGCNA? Should I use VST-transformed counts, TPMs, FPKMs, or something else if starting from FASTQ files? My plan is to first generate raw counts using nf-core/rnaseq, then apply VST transformation in DESeq2 before WGCNA. Is this okay?

Sample inclusion: If I have both healthy controls and disease samples, should I include all samples or only disease samples? I’ve read that WGCNA doesn’t require controls, but I’ve also seen suggestions that a reference group can help. I’m planning to combine datasets. While controls are age- and sex-matched within each dataset, there's still variability—especially across datasets. Should I limit WGCNA to disease samples only and apply batch correction?

Preprocessing pipeline: What’s the best pipeline/tool for local processing of FASTQ files for downstream WGCNA? I’m considering this pipeline: FastQC, fastp, HISAT2, featureCounts. Would you recommend this over using nf-core/rnaseq or GenPipes?

Use of fastp: Even if no adapters are detected, would you still recommend using fastp for consistency across datasets?

Thanks in advance!

[fchatonnet]

Dear tutorwhizz,

as mentioned here (https://www.biostars.org/p/280650/), WGCNA can be used with properly normalized RNA-seq data, apparently VST transformation is recommended. Some filtering over low expressed genes is also recommended, and some users even limit their analyses to DEG, but I would try both "DEG only" and "all filtered genes" strategies.
Then, if you want to identify biologically meaningful genes, I would include the control samples in your analysis since WGCNA clusterizes the genes with similar expression patterns. It will then hopefully output clusters of genes which behave differently between your control and disease states.
For pre-processing and primary analysis (basically QC, alignment and counts), I usually use fastQC, fastq_screen (to check for potential contamination by other organisms), cutadapt / trimgalore (although I usually don't perform any trimming, these tools are useful to remove adapters, over-represented sequences and polyA/G), samtools (to remove low quality reads), STAR and featureCounts. Everyone has their own preferred tools, so what you suggest is also fine by me. I would only strongly advise to use the exact same pipeline on all samples to avoid biases and batch effects (i.e. apply fastp or cutadapt even if no problematic read was detected in a particular sample). Do not forget to trim / clean both R1 and R2 fastq files at the same time if you use paired-end data, or the downstream aligner might fail!

Good luck!